Build a Meta Prompt That Forces a Model to Show Its Reasoning Steps

A confident answer with no visible reasoning is hard to check and easy to trust wrongly. This builds a standing meta-prompt you can drop in front of any task, forcing the model to restate the problem, name its assumptions, and reason in visible steps before it commits to a final answer.

<context>
You are a prompt design specialist who builds meta-prompts, prompts whose purpose is to shape how a model approaches other prompts. The user wants a reusable meta-prompt that makes a model expose its reasoning process before giving a final answer, for use on tasks where the reasoning matters as much as the answer.
</context>

<task>
**Generate** a meta-prompt that, when prepended to any task, makes the model:
1. Restate the task in its own words to confirm understanding
2. List the key assumptions it is making before answering
3. Work through its reasoning in visible, numbered steps
4. Flag any point where it is uncertain or where the task is ambiguous
5. Give a final answer clearly separated from the reasoning that produced it

The meta-prompt must work when prepended to an unrelated task, without needing to know what that task will be in advance.
</task>

<output_format>
- A single reusable block of instruction text, ready to prepend to any task
- No task-specific content inside it
- Include one short worked example showing the meta-prompt applied to a sample task
- Length: under 250 words for the meta-prompt itself
- Tone: precise, unambiguous, written as an instruction to a model
</output_format>